Author | Chen Wen

Source | Dongjian New Research Institute

On New Year's Eve in 2024, just before the CCTV Spring Festival Gala, OPPO held a "shortest press conference in history," where OPPO's Chief Product Officer Liu Zuohu announced, "OPPO officially enters the era of AI smartphones."

Shortly after the Chinese New Year holiday, Meizu publicly stated that it would cease production of "traditional smartphones" and transition to "All in AI."

From when the concept was first proposed to today, when almost all brands have intensively launched new AI smartphone products, it has only been half a year.

Especially after Apple released Apple Intelligence at WWDC, domestic smartphone brands have consistently targeted AI Agent as their direction for innovation.

Following this trend, a very obvious change in the smartphone industry is that while the focus of competition used to be on various parameter configurations and chip imaging capabilities, it is now taken over by AI Agent.

01 Apple Follows, Yet Doesn't Quite Follow

In contrast to its slow action on foldable screens, Apple has been very active in pursuing AI.

At WWDC in June, Apple launched the AI Agent-like product Apple Intelligence, which can help users proofread text, provide emergency email summaries, prioritize notifications, generate creative images, and more.

To make it easier to access AI functions, the iPhone 16 series, released in September, even includes a dedicated physical button. When Apple pushed out the iOS 18.1 update on October 29, Apple Intelligence officially went live. At the time, Tim Cook excitedly stated, "This will be the beginning of a new era."

However, unfortunately, due to regulatory policies, the launch of Apple Intelligence in the Chinese market is still uncertain.

When Tim Cook visited Sina Weibo headquarters on October 23 and was asked when Apple Intelligence would enter China, he responded, "We are working hard to advance, and there is a very specific regulatory process behind it that we need to complete. We also hope to bring it to Chinese consumers as soon as possible."

In fact, when OPPO released the Find X7 series in January this year, it had already introduced a batch of AI functions. The new phones are connected to a 7 billion-parameter large model on the device side, supporting nearly 100 AI functions such as AIGC removal, AI call summaries, and AI painters.

When Huawei released the Pura70 series in April, it upgraded its smart assistant to "Xiaoyi Intelligent Agent." This marked the beginning of the formation of the AI Agent concept in the smartphone industry.

Taking advantage of the gap before Apple Intelligence enters the Chinese market, Chinese smartphone brands have also accelerated the rollout of AI Agents.

On the evening of October 29, the same day Apple pushed out the iOS 18.1 update, Xiaomi released the Xiaomi 15 series equipped with the Pengpai OS 2 system. The famous "Xiaoai" assistant was upgraded to "Super Xiaoai," positioned as an AI assistant that can help you remember, find, and execute tasks. Xiaomi calls it a full-ecosystem AI smart assistant.

The next day (October 30), Honor released the Honor Magic 7, its first smartphone equipped with an AI Agent. The AI operating system, MagicOS 9.0, released earlier, is also integrated into the Magic 7. The YOYO assistant has been upgraded to YOYO Intelligent Agent.

The newly launched YOYO Intelligent Agent on the Honor Magic 7 possesses capabilities such as fuzzy understanding, interface recognition, automatic execution, and one-command execution. It can support various modes such as single-command system-level task execution, third-party application task execution, and even multi-application coordinated execution.

At the press conference, Honor CEO Zhao Ming demonstrated on-site how to automatically order 2,000 cups of Luckin Coffee using the Magic 7 and gave them to the attendees. Zhao Ming summarized it as "just a matter of one AI command."

In contrast to the rapid advancements made by Chinese smartphones in AI Agents, Apple Intelligence, which has not yet been launched in the Chinese market, is not as user-friendly as expected.

According to Bloomberg, internal Apple employees believe the company is at least two years behind its competitors in AI development.

Internal Apple research also shows that ChatGPT has a 25% higher accuracy rate and can answer 30% more types of questions compared to Siri, which is powered by Apple AI technology.

02 For Sales and to Compete for New Traffic Entrances

From the perspective of technological development trends, the integration of smartphones and large models is an inevitable outcome.

On the one hand, as former Alibaba CEO Daniel Zhang said, "All products deserve to be upgraded with large models." On the other hand, as a smart terminal that contains a large amount of users' private data, habits, and preferences and permeates life and work, the smartphone itself is also a very suitable platform for the implementation of generative AI.

Although there is currently no clear division between AI smartphones and traditional smartphones, making them two distinct categories, the changes brought by AI to smartphones are happening in reality.

Zhou Wei, Vice President of vivo, Vice President of OS Products, and Dean of vivo AI Global Research Institute, shared his and vivo's understanding of AI, stating, "AI's empowerment on smartphones only enhances the original user experience. People's demand for smartphones has not changed; large models just make interactions more natural and intuitive."

Li Jie, President of OnePlus China, also stated in a media interview, "The impact of AI on the smartphone industry will not come as a sudden change at a specific moment. This impact is gradual rather than explosive. For example, after three years, you may find that you no longer want to use a phone from three years ago. 'Lack of AI experience' will be one of the factors, especially when AI begins to change user habits or is deeply embedded in a certain usage scenario."

Thus, it is easy to understand that, similar to past competitions among manufacturers in systems, chips, and imaging functions, enhancing the product experience through AI Agents ultimately aims to return to the origin of "selling more smartphones."

IDC predicts that shipments of generative AI smartphones will increase by 364% year-on-year in 2024, reaching 234.2 million units. By 2028, global shipments of generative AI smartphones will reach 912 million units, with a compound annual growth rate of 78.4% from 2024 to 2028.

Just like foldable screens, such a large incremental market is not something any smartphone manufacturer would easily give up.

Besides maintaining product competitiveness, the deeper reason for smartphone manufacturers to engage in the competition of AI Agents lies in the FOMO (Fear of Missing Out) regarding new business models.

Currently, when we access services through our phones, we mostly use various independent apps, such as Meituan and Ele.me for food delivery, Didi and Gaode Maps for ride-hailing, and Taobao, JD.com, and Pinduoduo for shopping.

The emergence of AI Agents has turned the entry points for these services directly into smart assistants/agents like Siri, Xiaoai, and YOYO.

A typical scenario is when Zhao Ming purchased 2,000 cups of Luckin Coffee through YOYO at the press conference. On which platform was this service request placed? Meituan? Ele.me? Or the Luckin Coffee official website?

It is evident that in front of AI Agents, different food delivery platforms will compete fiercely to win this order. In this process, a new business rule will inevitably emerge to "arbitrate" the competition among food delivery platforms, and this new business rule is where smartphone manufacturers can generate incremental revenue.

03 The Focus of Competition Lies in the Device Side and Chips

Returning to the reality of smartphone competition, when everyone has an AI Agent, how can we differentiate the best from the rest?

We know that the capability of large models is proportional to the scale of computing power. This means that AI smartphones still need to return to the competition of model scale and computing power.

A very real problem is that due to the large amount of personal privacy stored on smartphones, many users are very resistant to uploading their data to the cloud and using cloud computing to solve computing power issues.

In June this year, after Apple officially announced its partnership with OpenAI in AI, Elon Musk immediately stated, "If Apple dares to integrate OpenAI into its operating system, I will ban Apple from all my companies. This is an unacceptable security violation."

Zhao Ming also stated, "Due to the need to protect user privacy and security, without access to a user's work history over the past decade, large cloud-side models cannot generate reports. Therefore, to truly help users better manage their affairs, we still need to rely on device-side AI capabilities."

In fact, not only considering the direction of user data leakage, compared to cloud-side large models, device-side large models also have significant cost advantages.

Cloud-side large models require the computing power support of cloud vendors, resulting in high costs. For example, vivo has revealed that if its 300 million users invoke a large model 10 times a day, the daily computing cost would be as high as 30 million yuan, with an annual average cost of 9-10 billion yuan.

From a technical perspective, the biggest challenge for smartphone manufacturers to implement large models in smartphones under limited hardware conditions lies in the "transformation" of lightweight device-side large models, which involves compressing and optimizing the model to reduce computational and storage requirements without affecting performance.

Once the model algorithm issue is resolved, the next comparison for AI Agents will be the size of computing power, returning to the old path of chip competition.

We can observe that since the beginning of this year, both smartphone and chip manufacturers have consistently leaned towards AI when launching new products, striving to meet the increasing demand for AI computing power as much as possible.

For example, the A18 Pro chip, released along with the iPhone 16 series, adopts a 16-core neural engine with a computing power of 35 TOPS (or 35 trillion operations per second). Compared to the A11 chip with a dual-core design, its computing power has increased by almost 58 times.

Similarly, closely following the A18 Pro, MediaTek released the Dimensity 9400, which adopts a second-generation full large-core architecture and also focuses on enhancing AI capabilities. It is equipped with MediaTek's new eighth-generation AI processor NPU 890.

The Dimensity 9400 further improves long-text understanding capabilities and AI model text support length on the device side. It also supports a large model running speed of 50 tokens per second, plus support for multimodal AI models, enabling smartphones to have more AI application scenarios.

In addition, Qualcomm's newly released Snapdragon 8 Ultimate Edition also adopts a full large-core structure, including two Phoenix L ultra-large cores with a maximum clock speed of 4.32GHz and six Phoenix M large cores with a maximum clock speed of 3.53GHz. Its single-core performance is improved by 45%, and its multi-core performance is also improved by 45%. Since the Snapdragon 8 Ultimate Edition integrates the Hexagon NPU, its AI performance is also enhanced by 45%.

04 Conclusion

Whether it's "shrinking" large models or enhancing chip computing power, both need to find their own balance. The former concerns the intelligence level of intelligent agents, while the latter is related to power consumption. Connecting these two parts constitutes a key factor in determining the quality of the user experience.

It becomes clear that if smartphone manufacturers want to remain competitive in the AI era, they must score highly in solving the impossible triangle of "model lightweighting - high computing power and low power consumption - user experience."